Gemini 4 Carbon

Google says Carbon is its strongest model, aimed at long-horizon agentic work, complex multi-step coding and scientific research, and that it uses more tokens than Gemini 4 Argon to reason deeply. It is closed weights with a context of 1,048,576 tokens. The list price is $2 in and $10 out per million tokens, the same introductory price as Argon. Google also says it ranks #1 on CursorBench 4.0.

The HN thread has 26 points. byefruit wrote "Looks like it's Ultra only." and says this is their last month on Gemini Pro. dom96 ran it on their own benchmark and says "it matches Fable 5.2", but "it is still slower and more costly than Fable 5.2". On X, Da7_Tech ran it in Devin for under $4 and wrote that "the sea is really bad".

All six prompts worked on the first try, 18 min 40 s and $1.05 in total at list price. The pelican took 3 min 2 s against a median of 75 s over 50 models. The About text took 33 s against a median of 8 s. The pelican wears a sailor cap and red scarf, with a fish in the basket. The Breakout game loads on a neon-styled start screen that reads click or press space to start.